Model Introduction

Doubao Speech Synthesis Large Model 2.0 (Seed-TTS 2.0) is a new-generation large-scale autoregressive speech synthesis foundation model developed by ByteDance. Trained on massive multilingual speech data and the semantic understanding capabilities of large language models, it breaks through the traditional TTS paradigm of one-way "text-to-speech" mapping and achieves semantic-aware emotionalized speech generation. It reaches industry-leading levels in naturalness, human-likeness, and controllability, and can be widely applied in human-computer interaction, content production, and multilingual localization scenarios.

Basic Technical Specifications

Item	Description
Model identifier	Platform Model ID: seed-tts-2.0
Audio sample rate	24kHz HD sampling
Output format	Supports mainstream formats such as mp3, wav, pcm, ogg_opus
Access method	HTTP unidirectional streaming interface

Billing

Doubao TTS 2.0 uses a per-character billing model:

• Unit price: 3 RMB / 10,000 characters
• Billing unit: character. 1 Chinese character, letter, punctuation mark, space, or line break counts as 1 character
• Pre-freeze basis: before the call, the quota is pre-frozen based on the character count of text in the request body
• Settlement basis: settled based on the actual character count returned by the upstream (usage.text_words); any over-frozen amount is automatically released

Note

• Only successfully synthesized requests are billed. Requests that fail due to content review or other reasons are not charged.
• The actual billed character count is based on the usage.text_words field returned by the interface.

Application Scenarios

• Audio/Video Creation & Editing: for short videos, ads, and e-commerce voice-overs, you can quickly create a large number of voices, allowing users to easily create voices that fit their videos, with flexible control over synthesis effects (style, emotion, local modifications), enabling one-stop generation of speech, background music, and sound effects.
• Audiobooks: for the production of audiobooks, multi-cast narration, and audio dramas, it can replace human voice actors, significantly reducing manual recording costs and quickly producing high-quality audio content.
• Game Voiceover: replaces some human recordings, adapting to multi-character, multi-emotion game scenarios with stronger controllability, and supports one-stop generation of speech, background music, and sound effects.
• AI Podcast Production: quickly generates multi-character dialogue audio, lowering production thresholds and enabling scaled, batched podcast content output.